claw-mem v2.11.0

claw-mem v2.11.0 - Benchmark Evaluation Support

Based on Paper

• "Memory for Autonomous LLM Agents" (2026)

New Features

Benchmark Support

• MemoryArena: Multi-session cross-domain tasks (web navigation, preference planning, progressive search, formal reasoning)
• MemBench: 4D capability evaluation (retrieval accuracy, test-time learning, long-range understanding, selective forgetting)
• EvoMemory: Streaming task evaluation for self-evolving memory

Unified Runner

• BenchmarkRunner: Run all benchmarks with single command
• BenchmarkReport: Structured results with visualization support

Files

• src/claw_mem/benchmarks/ (new module)
• tests/test_benchmarks.py (25 tests)

Testing

• 25 tests passed

Reference

• Paper: Memory for Autonomous LLM Agents (2026)
• MemoryArena: arXiv:2602.16313
• MemBench: arXiv:2506.21605